The elusiveness of drug interactions

One of the seemingly insurmountable challenges that face CPOE, pharmacy, and e-Prescribing is alert fatigue. Alert fatigue results when something like drug drug interactions (DDIs) are highly insensitive or have a high rate of false positives. When a clinical information system imports DDIs from a commerical provider like FirstDataBank (FDB), you get every single possible interaction that may occur. This paradoxically results in a situation where noisy alerting may impede clinical decision support rather than facilitate it. In Greenberg and Ridgely's JAMIA commentary, Clinical Decision Support and Malpractice Risk, they poignantly highlight:

"CDS represents a situation in which malpractice and products liability can too easily lead to a perverse equilibrium in which the law has a detrimental effect on technology and in which patients, physicians, institutions, and the government are all made worse off as a result."

Once you understand the current state, we want to focus on "where do we go from here?" Philosophically, there are few steps:

1. Construct a De facto standard for DDIs. Ideally, this would be an expert consensus list containing the most pertinent and clinically relevant interactions. Currently, the RAND corporation is working with Partners Health Care to achieve this.
2. If a consensus DDI list existed, regulators like the ONC, CMS, and Joint Commision could endorse the list and say something to the effect of: "If you choose to just use this DDI list as a minimum, that would satisfy our standards."
3. This may eventually create a "liability shelter" where if someone chose to use the expert consensus list instead of a huge all inclusive list, it could not be used as a basis of malpractice or as evidence of negligence.

Ultimately, the informatics community needs to address this issue by setting standards proactively rather than let the standards be set for them through tort and laywers after the fact.

At our hospital, we are trying to refine our FDB interactions through a clinical vetting process. We meet weekly and generate a SQL report of the most frequent contraindicated interactions that fired the past week. We then refine the DDI pairs into 3 categories.

1. Absolutely contraindicated per the package insert with no clinical exception.
2. Absolutely contraindicated per the package insert with clinical exceptions and uses.
3. Erroneously classified as contraindicated by FDB. We then reassign it the appropriate severity.
4. Rinse and repeat with severe interactions etc.

It also helps if a CPOE or pharmacy information system vendor offers the granular ability to customize the way DDIs are displayed. In the past, we have tried to attach SQL queries to DDI pairs to address clinical need.

EXAMPLE:

Since FDB does not factor drug doses in interaction pairs, we could not address the recent FDA update of limiting simvastatin doses to 10mg while a patient is on verapamil. The interaction could be categorized as "severe", but that does not offer enough clinical guidance. To solve this, a SQL query was attached and written to the effect of:

select distinct tprx.genericmedid

from table_patient.rx tprx

where tprx.genericmedid='simvastatinID' AND tprx.dose > 10

This means that the severe simvastatin-verapamil interaction would only fire if the simvastatin dose ordered was greater than 10 mg. Using this workaround, we also wrote a similiar query to trigger alerts based on age to differentiate adult vs ped patients. Overall this has worked well but only serves as a band-aid until DDIs and alert fatigue are worked out in a more elegant (maybe national) solution.